The English edition is being prepared.

OpenAI Tightens AI Training Security—and Explains Why Stronger Models Need Stronger Boundaries